Edd Barrett wrote: > I notice that performance of learning spam with the sqlite backend is > pretty poor (can take over a minute to learn a message if you leave it > learning a corpus long enough -- during which the hard disk LED is > pegged on). Upstream speculates that fsync() may be expensive on > OpenBSD, which sqlite apparently uses liberally(?) Anyone know anything > about that?
Well, it does one or two fsync per transaction. if you're doing a thousand auto-commit updates, it's gonna be slow.
